io.github.ralfbecher/orionbelt-semantic-layer

Ask an LLM to write SQL against a raw star schema and sooner or later it joins two fact tables and hands you a revenue number inflated by a factor of eight. It looks right. Nobody catches it.

OrionBelt is a semantic sidecar. You declare dimensions, measures, metrics, and joins in version-controlled YAML. OrionBelt compiles them into dialect-specific SQL through a real AST, and routes multi-fact queries through a Composite Fact Layer planner that blocks the join paths that produce fan traps. Agents and BI tools ask for "Total Revenue" by "Country". They never see a table name.

No BI tool in the middle. No runtime lock-in. Point it at what you already have.

Here is TPC-DS query 98. Two measures over the same column, identical but for one line: Class Revenue is pinned to a coarser grain than the query asks for.

measures:
  Store Sales Amount:
    columns: [{dataObject: Store Sales, column: Ext Sales Price}]
    aggregation: sum

  Class Revenue:
    columns: [{dataObject: Store Sales, column: Ext Sales Price}]
    aggregation: sum
    grain: {mode: FIXED, keepOnly: [Class]}   # <- pin to Class, ignore query grain

metrics:
  Revenue Ratio:
    expression: "{[Store Sales Amount]} * 100.0 / {[Class Revenue]}"

That one grain line is what becomes SUM(...) OVER (PARTITION BY "Class") below.

The query names business concepts. No tables, no joins, no SQL:

select:
  dimensions: [Item ID, Item Description, Category, Class, Current Price]
  measures: [Store Sales Amount, Revenue Ratio]
where:
  - {field: Category, op: inlist, value: [Sports, Books, Home]}
  - {field: Order Date, op: between, value: ["1999-02-22", "1999-03-24"]}

pip install orionbelt-semantic-layer
obsl compile tpcds.obml.yml -q Q98.yml -d duckdb

WITH "base" AS (
  SELECT
    "Item"."i_item_id" AS "Item ID",
    "Item"."i_item_desc" AS "Item Description",
    "Item"."i_category" AS "Category",
    "Item"."i_class" AS "Class",
    "Item"."i_current_price" AS "Current Price",
    CAST(SUM("Store Sales"."ss_ext_sales_price") AS DECIMAL(18, 2)) AS "Store Sales Amount",
    SUM("Store Sales"."ss_ext_sales_price") AS "Class Revenue"
  FROM "main"."store_sales" AS "Store Sales"
  LEFT JOIN "main"."item" AS "Item"
    ON "Store Sales"."ss_item_sk" = "Item"."i_item_sk"
  LEFT JOIN "main"."date_dim" AS "Date"
    ON "Store Sales"."ss_sold_date_sk" = "Date"."d_date_sk"
  WHERE
    "Item"."i_category" IN ('Sports', 'Books', 'Home')
    AND "Date"."d_date" BETWEEN '1999-02-22' AND '1999-03-24'
  GROUP BY ALL
)
SELECT
  "Item ID" AS "Item ID",
  "Item Description" AS "Item Description",
  "Category" AS "Category",
  "Class" AS "Class",
  "Current Price" AS "Current Price",
  "Store Sales Amount" AS "Store Sales Amount",
  "Store Sales Amount" * 100.0 / NULLIF(SUM("Class Revenue") OVER (PARTITION BY "Class"), 0) AS "Revenue Ratio"
FROM "base" AS "base"
ORDER BY
  "Category" ASC,
  "Class" ASC,
  "Item ID" ASC,
  "Item Description" ASC,
  "Revenue Ratio" ASC

You did not write the join path, the window function over an aggregate, the NULLIF guard, or one table name. Change -d duckdb to -d snowflake and the same two files compile for Snowflake, or for any of eight dialects.

This is checked, not asserted. 40 TPC-DS queries are built against a single OBML model and compared row by row against each engine's own reference SQL: 39 of 40 match on DuckDB at sf=1, 37 of 40 on ClickHouse at sf=10. Every one of the remaining differences traces to a reference variant rather than a compilation error, and each is documented. See the sweep, or the queries in examples/tpcds_queries/.

The same model serves every surface you already use:

• Your BI tool, over the PostgreSQL wire protocol on :5432. Tableau, Power BI, Superset, DBeaver, and psql connect with the Postgres driver they already ship. Dremio federates it as a Postgres source.
• Your AI agents, over MCP. Works with Claude, Cursor, Copilot, and Windsurf.
• Your code, over REST, Arrow Flight SQL, or PEP 249 drivers.

Compiles to BigQuery, ClickHouse, Databricks, Dremio, DuckDB/MotherDuck, MySQL, PostgreSQL, and Snowflake.

Where OrionBelt fits

OrionBelt is a sidecar, not a platform. It compiles a YAML model into correct SQL and exposes it over the protocols you already use. It does not run a cluster, own your cache, or ask you to adopt a cloud.

Reach for OrionBelt when:

• Agents query your data and a silently wrong number is unacceptable. Multi-fact queries route through a Composite Fact Layer planner that blocks fan-trap join paths instead of quietly summing across them.
• You want your metric definitions in reviewable YAML, with no JavaScript or Python in the model layer.
• Your BI tool should connect over the Postgres driver it already ships, with no new connector to install and no vendor runtime in the path.
• You self-host, across more than one engine, and want one model to compile for all of them.

Reach for something else when:

• You need pre-aggregation and caching tuned for high-concurrency dashboards at scale. Cube has years of production hardening there that OrionBelt does not.
• Your metrics already live in dbt and your team is happy there. MetricFlow keeps them where they are.
• You want an exploratory analysis language rather than a serving layer. Malloy is a better fit.

Try the live demo with a pre-loaded model, or open the Colab notebook and run it against TPC-H data.

Contents

Try it in 30 seconds · Claude Desktop / MCP · Why OrionBelt? · Features · Example · Documentation · Roadmap · Commercial · Development

Try it in 30 Seconds

Option A: Live Demo (no install)

Open the Live Demo — Gradio UI with a pre-loaded example model. Paste a query, pick a dialect, see SQL instantly.

API explorer: Swagger UI | ReDoc

Want to try the PostgreSQL wire surface? Cloud Run is HTTPS-only, so the public demo can't expose ports 5432 (pgwire) or 8815 (Flight SQL). Spin the same demo up locally in two commands — it includes the baked-in orionbelt_1_commerce DuckDB dataset and the full OBSQL surface:

bash

docker run --rm -d --name orionbelt-demo \
  -p 8080:8080 -p 5432:5432 -p 8815:8815 \
  -e PGWIRE_ENABLED=true \
  -e FLIGHT_ENABLED=true \
  ralforion/orionbelt-semantic-layer-api:latest

# REST + Gradio UI:   http://localhost:8080/ui
# pgwire (any psql / DBeaver / Tableau / Power BI):
psql "host=localhost port=5432 user=obsl dbname=orionbelt_1_commerce sslmode=disable" \
  -c 'SELECT "Client Name", "Total Sales" LIMIT 5'
# Flight SQL smoke test:
uv run python examples/obsql.py 'SELECT "Client Name", "Total Sales" LIMIT 5'

docker stop orionbelt-demo

The container ships with PGWIRE_AUTH_MODE=trust (default), so it's safe for localhost but not safe to expose to the public internet. For exposed deployments, set AUTH_MODE=api_key (shipped in v2.12.0): pgwire then negotiates SCRAM-SHA-256 (or cleartext over TLS) against the shared key store.

Option B: Google Colab (no install)

— Interactive notebook with TPC-H data: explore the model, compile queries across dialects, execute against DuckDB, and see results. Requires Python 3.12 runtime.

Option C: Install from PyPI

pip install orionbelt-semantic-layer

Then paste into a Python REPL:

from orionbelt.parser import ReferenceResolver, TrackedLoader
from orionbelt.compiler.pipeline import CompilationPipeline
from orionbelt.models.query import QueryObject, QuerySelect

model_yaml = """
version: 1.0
dataObjects:
  Orders:
    code: ORDERS
    columns:
      Price: { code: PRICE, abstractType: float }
      Country: { code: COUNTRY, abstractType: string }
dimensions:
  Country:
    dataObject: Orders
    column: Country
    resultType: string
measures:
  Total Revenue:
    resultType: float
    aggregation: sum
    expression: "{[Orders].[Price]}"
"""

loader = TrackedLoader()
raw, source_map = loader.load_string(model_yaml)
resolver = ReferenceResolver()
model, result = resolver.resolve(raw, source_map)

query = QueryObject(select=QuerySelect(dimensions=["Country"], measures=["Total Revenue"]))
pipeline = CompilationPipeline()
output = pipeline.compile(query, model, "postgres")
print(output.sql)

Output:

SELECT
  "Orders"."COUNTRY" AS "Country",
  CAST(SUM("Orders"."PRICE") AS NUMERIC(18, 2)) AS "Total Revenue"
FROM ORDERS AS "Orders"
GROUP BY "Orders"."COUNTRY"

No env file needed — the compilation pipeline is stateless.

Start the servers:

orionbelt-api                              # REST API on :8000 (Swagger UI at /docs, Gradio UI at /ui)
orionbelt-ui                               # standalone Gradio UI on :7860 (connects to API on :8000)
FLIGHT_ENABLED=true orionbelt-api          # API + Arrow Flight SQL on :8815 (DBeaver, Tableau, Power BI)
PGWIRE_ENABLED=true orionbelt-api          # API + PostgreSQL wire on :5432 (Tableau, DBeaver, Superset, psql, Dremio source)

Option C2: Install with uv

uv pip install orionbelt-semantic-layer

uv run orionbelt-api                       # REST API on :8000 (Swagger UI at /docs, Gradio UI at /ui)
uv run orionbelt-ui                        # standalone Gradio UI on :7860 (connects to API on :8000)
FLIGHT_ENABLED=true uv run orionbelt-api   # API + Arrow Flight SQL on :8815 (DBeaver, Tableau, Power BI)
PGWIRE_ENABLED=true uv run orionbelt-api   # API + PostgreSQL wire on :5432 (Tableau, DBeaver, Superset, psql, Dremio source)

Use the obsl CLI (no server needed - compiles in-process):

obsl validate model.yaml                                  # lint a model (exit 1 on error, CI-friendly)
obsl compile model.yaml -q query.json -d snowflake        # print the generated SQL
obsl compile model.yaml --sql 'SELECT "Region", "Sales" FROM model'  # ... or from an OBSQL string
obsl describe model.yaml                                   # overview of data objects + artefacts
obsl diagram model.yaml                                    # Mermaid ER diagram
obsl convert obml-to-osi model.yaml                        # OBML -> OSI (and osi-to-obml)
obsl execute -q query.json --server http://host           # run against a deployed model (omit MODEL)

See the CLI guide for all commands.

Smoke-test the Flight SQL surface without a BI tool:

uv run python examples/obsql.py 'SELECT version()'
uv run python examples/obsql.py 'SHOW TABLES'
uv run python examples/obsql.py 'SELECT "Region", "Total Sales" FROM sales LIMIT 5'

# Multi-model deployment? Pick the model with -m:
uv run python examples/obsql.py -m sales 'SHOW TABLES'
uv run python examples/obsql.py --list   # discover loaded models via REST

Try OBSQL in 30 seconds

OBSQL — OrionBelt Semantic QL — is the SQL surface BI tools and humans actually write. Bare labels, MEASURE() markers, or matching aggregate wrappers; aggregation-match validation; WITH ROLLUP / WITH CUBE; no escape hatch to raw warehouse SQL. Same language over Arrow Flight SQL (v2.4+) and PostgreSQL wire (v2.5+):

PGWIRE_ENABLED=true uv run orionbelt-api &

# Every BI tool already ships a Postgres ODBC/JDBC driver — point yours at :5432
psql "host=localhost port=5432 user=obsl dbname=sales sslmode=disable" \
  -c 'SELECT "Region", "Total Sales" LIMIT 5'

# All three measure forms compile to the same vendor SQL:
psql "..." -c 'SELECT "Region", "Total Sales"        FROM sales LIMIT 5'  -- bare
psql "..." -c 'SELECT "Region", MEASURE("Total Sales") FROM sales LIMIT 5'  -- explicit marker
psql "..." -c 'SELECT "Region", SUM("Total Sales")   FROM sales LIMIT 5'  -- matching aggregate

See the OBSQL reference for the full grammar.

Option D: Docker

Stage 1 — Zero-config start (models loaded later via API or UI):

docker run -p 8080:8080 ralforion/orionbelt-semantic-layer-api

Open http://localhost:8080/docs to explore the API.

Stage 2 — Realistic setup with docker compose:

# docker-compose.yml
services:
  api:
    image: ralforion/orionbelt-semantic-layer-api:2.25.1
    ports: ["8080:8080"]
    env_file: .env
    volumes:
      - ./models:/app/models:ro
    environment:
      MODEL_FILES: /app/models/my-model.obml.yml

  ui:
    image: ralforion/orionbelt-semantic-layer-ui:2.25.1
    ports: ["7860:7860"]
    environment:
      API_BASE_URL: http://api:8080

docker compose up -d

See .env.template for the full environment variable reference.

Docker notes:

• API_SERVER_HOST is already 0.0.0.0 inside the container — no override needed.
• MCP via stdio does not work in Docker. Use the MCP HTTP client for containerized deployments.
• Mount models to /app/models (or any path) and set MODEL_FILES (comma-separated paths) to pre-load on startup.
• For production, pin a version tag (:2.25.1) rather than :latest.

Claude Desktop / MCP

The MCP server is a separate thin client that delegates to the REST API:

orionbelt-semantic-layer-mcp

Add to your Claude Desktop claude_desktop_config.json:

{
  "mcpServers": {
    "orionbelt": {
      "command": "uvx",
      "args": ["orionbelt-semantic-layer-mcp"]
    }
  }
}

Also works with Copilot, Cursor, and Windsurf. See the MCP repo for full setup options.

Why OrionBelt?

Features

Semantic Modeling

• OBML Format — YAML-based semantic models with data objects, dimensions, measures, metrics, and joins
• Cross-Schema Queries — model data objects across multiple databases and schemas in a single model
• Static Model Filters — mandatory WHERE conditions baked into the model, auto-applied with join extension
• OBSL Graph & SPARQL — RDF graph export and read-only SPARQL querying for every loaded model
• OSI Interoperability — bidirectional conversion between OBML and the Open Semantic Interchange format, now developed as Apache Ossie (incubating)

SQL Compilation

• 8 SQL Dialects — BigQuery, ClickHouse, Databricks, Dremio, DuckDB/MotherDuck, MySQL, Postgres, Snowflake
• AST-Based Generation — custom SQL AST ensures correct, injection-safe SQL (not string templates)
• Star Schema & CFL — automatic join resolution with Composite Fact Layer for multi-fact queries
• Data Types & Precision — automatic CAST wrapping with dialect-specific type rendering and precision clamping
• Display Formatting — number format patterns (#,##0.00, 0.00%) on measures/metrics with locale-aware rendering
• Timezone Settings — auto-detect database session timezone with defaultTimezone fallback and ISO 8601 serialization
• sqlglot Validation — post-generation syntax check across all supported dialects

Integration Surface

• REST API — FastAPI endpoints for model management, validation, compilation, and execution
• MCP Server — separate thin client for Claude, Copilot, Cursor, Windsurf
• AI Integrations — LangChain, OpenAI Agents SDK, CrewAI, Google ADK, Vercel AI SDK, n8n, ChatGPT
• Gradio UI — interactive web interface for model editing, query testing, and ER diagrams
• DB-API 2.0 + Flight SQL — PEP 249 drivers and Arrow Flight SQL server for DBeaver, Tableau, Power BI; ships with examples/obsql.py, a tiny terminal CLI for testing the Flight surface without a BI tool
• PostgreSQL Wire Protocol (v2.5.0+) — native Postgres-protocol surface on :5432. Every BI tool already ships a Postgres ODBC/JDBC driver, so the user side is "point your existing connection at OBSL and go" — Tableau, DBeaver, Superset, Power BI, plain psql, and Dremio as a federated Postgres source (Dremio → OBSL → optionally back to Dremio's lakehouse, full circle)

Agent-Facing API

• Model Health on Load — every model load returns a health block with orphan dataObjects, fan-trap risks, and unreachable dimensions — agents skip the defensive second round trip
• Query Plan Endpoint — POST /query/plan returns the planner's understanding (planner choice, physical tables, join path, would_compile) without compiling SQL or executing; opt-in include_database_explain adds the warehouse's raw EXPLAIN
• Structured Warnings — every warnings list across the API uses a stable {code, severity, message, path, hint, context} shape with a documented code taxonomy; agents branch on codes instead of parsing messages
• Fuzzy /find Recovery — when a search produces no exact or synonym hits, deterministic Levenshtein + trigram fallback returns near-miss candidates with scores and reasons
• Model Examples — optional OBML examples: block of canonical queries; GET /examples (with ?intent= filtering) gives agents one-round-trip discovery of what a model is designed to answer

Freshness-Driven Result Cache

• Source-level freshness contracts — declare refresh: blocks on dataObject entries (interval / heartbeat / static); the cache derives query TTLs from the contracts of the physical tables a query touched, not from caller guesses
• Heartbeat invalidation — one POST /v1/heartbeat to a physical table invalidates every cached query that depends on it, across every dataObject and session
• DuckDB metadata + Parquet results — file-backed cache with type-precise serialization, lazy expiration, LRU capacity sweep; opt-in via CACHE_BACKEND=file
• Inverts the Cube/dbt/Looker pattern — contracts live on the source, not the semantic abstraction; one source of truth across every cube/explore/saved query reading the table

Developer Experience

• Source-Position Errors — validation errors report exact YAML line and column
• ER Diagrams — interactive Mermaid diagrams with zoom and download (MD/PNG/Turtle)
• Session Management — TTL-scoped sessions with thread-safe model isolation
• JSON Schema — full OBML and query schema for IDE autocompletion (yaml-language-server)

Example

Define a Semantic Model (OBML)

# yaml-language-server: $schema=https://raw.githubusercontent.com/ralforion/orionbelt-semantic-layer/main/schema/obml-schema.json
version: 1.0
dataObjects:
  Customers:
    code: CUSTOMERS
    database: WAREHOUSE
    schema: PUBLIC
    columns:
      Customer ID: { code: CUSTOMER_ID, abstractType: string }
      Country:     { code: COUNTRY, abstractType: string }

  Orders:
    code: ORDERS
    database: WAREHOUSE
    schema: PUBLIC
    columns:
      Order Customer ID: { code: CUSTOMER_ID, abstractType: string }
      Price:             { code: PRICE, abstractType: float }
      Quantity:          { code: QUANTITY, abstractType: int }
    joins:
      - joinType: many-to-one
        joinTo: Customers
        columnsFrom: [Order Customer ID]
        columnsTo: [Customer ID]

dimensions:
  Country:
    dataObject: Customers
    column: Country
    resultType: string

measures:
  Revenue:
    resultType: float
    aggregation: sum
    expression: "{[Orders].[Price]} * {[Orders].[Quantity]}"
    dataType: "decimal(18, 2)"

Compile via REST API

# Create a session
curl -s -X POST http://localhost:8080/v1/sessions | jq .session_id
# -> "a1b2c3d4"

# Load the model
curl -s -X POST http://localhost:8080/v1/sessions/a1b2c3d4/models \
  -H "Content-Type: application/json" \
  -d '{"model_yaml": "..."}' | jq .model_id
# -> "abcd1234"

# Compile a query
curl -s -X POST http://localhost:8080/v1/sessions/a1b2c3d4/query/sql \
  -H "Content-Type: application/json" \
  -d '{"model_id":"abcd1234","query":{"select":{"dimensions":["Country"],"measures":["Revenue"]}},"dialect":"postgres"}' \
  | jq -r .sql

SELECT
  "Customers"."COUNTRY" AS "Country",
  CAST(SUM("Orders"."PRICE" * "Orders"."QUANTITY") AS NUMERIC(18, 2)) AS "Revenue"
FROM WAREHOUSE.PUBLIC.ORDERS AS "Orders"
LEFT JOIN WAREHOUSE.PUBLIC.CUSTOMERS AS "Customers"
  ON "Orders"."CUSTOMER_ID" = "Customers"."CUSTOMER_ID"
GROUP BY "Customers"."COUNTRY"

Change dialect to bigquery, clickhouse, databricks, dremio, duckdb, mysql, or snowflake for dialect-specific SQL.

Gradio UI

• SQL Compiler — side-by-side OBML model and query editors with syntax highlighting, 8 dialect selector, one-click compilation with formatted SQL output and query explain
• Query Execution — execute compiled queries against a connected database, view results with locale-aware number formatting, response metadata panel, TSV download and clipboard copy (requires QUERY_EXECUTE=true)
• ER Diagram — interactive Mermaid ER diagram with zoom, column toggle, and download (MD/PNG/Turtle)
• Ontology Graph — interactive vis-network visualization of the OBML graph (data objects, dimensions, measures, metrics, joins) with toggleable layers and adjustable node spacing
• Editor Toolbar — clear, undo, redo, upload, download, and copy buttons on all code editors
• OSI Import/Export — convert between OBML and OSI formats
• Dark/Light Mode — toggle via header button, state persisted across sessions

Embedded mode — the UI is mounted at /ui on the API server:

pip install orionbelt-semantic-layer && orionbelt-api
# -> UI at http://localhost:8000/ui

Standalone mode — run API and UI as separate processes:

orionbelt-api                                              # API on :8000
orionbelt-ui                                               # UI on :7860 (connects to API on :8000)
API_BASE_URL=http://remote-api:8080 orionbelt-ui           # point UI to a remote API

Documentation

Status & Roadmap

Commercial Offerings

OrionBelt Semantic Layer is open by default — the OSS distribution has full parity on the shipped v2.6 surface and is production-grade for self-hosted use. For teams that want production support, a managed runtime, or embedded analytics terms, RALFORION offers:

• Embedded analytics license — relicensing terms for shipping OBSL inside a commercial product
• Commercial cloud offering — managed OrionBelt runtime with SLAs
• Enterprise features — capabilities tailored for enterprise deployments
• Consulting + support — implementation, modeling, and production support

Contact RALFORION d.o.o. for details.

Companion Project

OrionBelt Analytics

An ontology-based MCP server that analyzes relational database schemas and generates RDF/OWL ontologies. Together with OrionBelt Semantic Layer, it enables AI assistants to navigate your data landscape through ontologies and compile safe, dialect-aware analytical SQL.

Development

Contributing to OrionBelt or running from source:

git clone https://github.com/ralforion/orionbelt-semantic-layer.git
cd orionbelt-semantic-layer
uv sync                           # install all deps (dev, docs, ui, flight, drivers)
uv run orionbelt-api              # start API on :8000

# Quality
uv run pytest                     # run tests
uv run ruff check src/            # lint
uv run ruff format src/ tests/    # format
uv run mypy src/                  # type check

# Docs
uv sync --extra docs && uv run mkdocs serve  # docs on :8080

License

Copyright © 2026 RALFORION d.o.o.

OrionBelt® is a registered trademark of RALFORION d.o.o.

Licensed under the Business Source License 1.1. The Licensed Work will convert to Apache License 2.0 on 2030-03-16.

By contributing to this project, you agree to the Contributor License Agreement.

For commercial licensing inquiries, contact: licensing@ralforion.com